Inference Speed Evaluation on Commodity Hardware Inference Efficiency Benchmark
11.91Latency (ms)FusionProxy
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FusionProxyPlatform=RTX 4070, Resolution=480 × 640, Precision=FP16, Framework=PyTorch2026.05 | 11.91 | 83.98 | |
| FusionProxyPlatform=RTX 4070, Resolution=768 × 1024, Precision=FP16, Framework=PyTorch2026.05 | 19.32 | 51.75 | |
| FusionProxyPlatform=RTX 3060, Resolution=480 × 640, Precision=FP16, Framework=PyTorch2026.05 | 20.95 | 47.7 | |
| FusionProxyPlatform=Apple M3, Resolution=480 × 640, Precision=FP16, Framework=PyTorch2026.05 | 26.4 | 37.8 | |
| FusionProxyPlatform=RTX 3060, Resolution=768 × 1024, Precision=FP16, Framework=PyTorch2026.05 | 68.11 | 14.7 | |
| FusionProxyPlatform=Apple M3, Resolution=768 × 1024, Precision=FP16, Framework=PyTorch2026.05 | 82.5 | 12.1 |